Chapter Five Controllability and Observability

advertisement
Chapter Five
Controllability and Observability
Controllability and observability represent two major concepts of modern control
system theory. These originally theoretical concepts, introduced by R. Kalman
in 1960, are particularly important for practical implementations. They can be
roughly defined as follows.
Controllability: In order to be able to do whatever we want with the given
dynamic system under control input, the system must be controllable.
Observability: In order to see what is going on inside the system under
observation, the system must be observable.
Even though the concepts of controllability and observability are almost abstractly defined, we now intuitively understand their meanings. The remaining
problem is to produce some mathematical check up tests and to define controllability and observability more rigorously. Our intention is to reduce mathematical
derivations and the number of definitions, but at the same time to derive and
define very clearly both of them. In that respect, in Section 5.1, we start with
observability derivations for linear discrete-time invariant systems and give the
corresponding definition. The observability of linear discrete systems is very
naturally introduced using only elementary linear algebra. This approach will
be extended to continuous-time system observability, where the derivatives of
measurements (observations) have to be used, Section 5.2. Next, in Sections 5.3
and 5.4 we define controllability for both discrete- and continuous-time linear
systems.
In this chapter we show that the concepts of controllability and observability
are related to linear systems of algebraic equations. It is well known that a
221
222
CONTROLLABILITY AND OBSERVABILITY
solvable system of linear algebraic equations has a solution if and only if the rank
of the system matrix is full (see Appendix C). Observability and controllability
tests will be connected to the rank tests of ceratin matrices, known as the
controllability and observability matrices.
At the end of this chapter, in Section 5.5, we will introduce the concepts
of system stabilizability (detectability), which stand for controllability (observability) of unstable system modes. Also, we show that both controllability and
observability are invariant under nonsingular transformations. In addition, in the
same section the concepts of controllability and observability are clarified using
different canonical forms, where they become more obvious.
The study of observability is closely related to observer (estimator) design,
a simple, but extremely important technique used to construct another dynamic
system, the observer (estimator), which produces estimates of the system state
variables using information about the system inputs and outputs. The estimator
design is presented in Section 5.6. Techniques for constructing both full-order and
reduced-order estimators are considered. A corresponding problem to observer
design is the so-called pole placement problem. It can be shown that for a
controllable linear system, the system poles (eigenvalues) can be arbitrarily
located in the complex plane. Since this technique can be used for system linear
feedback stabilization and for controller design purposes, it will be independently
presented in Section 8.2.
Several examples are included in order to demonstrate procedures for examining system controllability and observability. All of them can be checked
by MATLAB. Finally, we have designed the corresponding laboratory experiment by using the MATLAB package, which can contribute to better and deeper
understanding of these important modern control concepts.
Chapter Objectives
This chapter introduces definitions of system controllability and observability. Testing controllability and observability is replaced by linear algebra problems of finding ranks of certain matrices known as the controllability and observability matrices. After mastering the above concepts and tests, students will
be able to determine system initial conditions from system output measurements,
under the assumption that the given system is observable. As the highlight of
this chapter, students will learn how to construct a system’s observer (estimator),
223
CONTROLLABILITY AND OBSERVABILITY
which for an observable system produces the estimates of state variables at any
time instant.
5.1 Observability of Discrete Systems
Consider a linear, time invariant, discrete-time system in the state space form
!#"$
(5.1)
with output measurements
% &')( &
(5.2)
where &+*-,/.0 % &+*-,21 . and ( are constant matrices of appropriate
dimensions. The natural question to be asked is: can we learn everything about
the dynamical behavior of the state space variables defined in (5.1) by using
only information from the output measurements (5.2). If we know , then the
recursion (5.1) apparently gives us complete knowledge about the state variables
at any discrete-time instant. Thus, the only thing that we have to determine from
the state measurements is the initial state vector &34 .
Since the 5 -dimensional vector 6 has 5 unknown components, it
Take
is expected that 5 measurements are sufficient to determine .
789:;:<:; 5>= in (5.1) and (5.2), i.e. generate the following sequence
% &?@( % A
?@(BCD@(BEF6
% HG?@( GD@( JIK')( ML &
(5.3)
..
.
% 5N= C)(BOD 5= ')(PE R. Q!S 6
or, in matrix form
TUU % 6 UU % U % 6G
WYXX [ O. 1O\^]_S
XX
X
V
Z
..
.
% 5N= ( YW XX [ E. 1`\6]9.
UU ( XX
U ( L X
a &
.
..
V
Z
( .9 QKS
TUU
(5.4)
224
CONTROLLABILITY AND OBSERVABILITY
We know from elementary linear algebra that the system of linear algebraic
equations with b unknowns, (5.4), has a unique solution if and only if the system
matrix has rank b . In this case we need
ghh
j k Yq rr
P
hh jPkElk rr
h
r
cd
ef jPkElmk
b
..
i
u
s
t
j k l7. nRk o0p
Thus, the initial condition
ability matrix, defined by
vxw
(5.5)
is completely determined if the so-called observ-
j k q rr ~ nE`€69n
hh j k l k rr
h j k l km r
ghh
yPz l k|{ j kO}
t
has rank
b
i
..
.
j k l n9k oKp
s
(5.6)
, that is
cdCef y
t
b
(5.7)
The previous derivations can be summarized in the following theorem.
Theorem 5.1 The linear discrete-time system (5.1) with measurements (5.2)
is observable if and only if the observability matrix (5.6) has rank equal to b .
A simple second-order example demonstrates the procedure for examining
the observability of linear discrete-time systems. More complex examples corresponding to real physical control systems will be considered in Sections 5.7
and 5.8.
Example 5.1: Consider the following system with measurements
‚„ƒ z …†‡C}
ƒ p z…†‡C}
ˆ
m
 z…}
‡
tŽ
t
‚‡
Š
‚Œƒ 6z …}
p
‹ ˆ ƒ m z …}Rˆ
‚Œƒ
D‰  ƒ p z&z&…‘…}R} ˆ
m
‰
225
CONTROLLABILITY AND OBSERVABILITY
The observability matrix for this second-order system is given by
’”“–•u—B˜
“›•6œ
ž
—B˜E™˜š
’

œ8ŸDš
’¤“
“¥

Since the rows of the matrix
are linearly independent, then ¡8¢£
,
i.e. the system under consideration is observable. Another way to test the
completeness of the rank of square matrices is to find their determinants. In
¦§E¨
this case
’©“ª$«“ ¬ Ÿ4­
®Œ¯°;° ¡
¢£ “±¥²“ 
³
Example 5.2: Consider a case of an unobservable system, which can be
obtained by slightly modifying Example 5.1. The corresponding system and
measurement matrices are given by
™F˜ “–•´ª¶œ µ
ª 
ª« šx·
—P˜ “¸ œ
D¹
The observability matrix is
’”“ • ª$œ º
so that
¡C¢£
’»“ œN¼©
ª  œ8Ÿ š
, and the system is unobservable.
³
5.2 Observability of Continuous Systems
A linear, time invariant, continuous system in the state space form was studied
in Chapter 3. For the purpose of studying its observability, we consider an
input-free system ½
¾¿&ÀAÁ “ ™ ¾?¿&ÀAÁ
·
¾¿&À&Â
Á “ ¾Ã “ ¯ ¢£¢!Ä
Å$¢
(5.8)
with the corresponding measurements
Æ
¿ŒÀJÁ “ — ¾D¿&ÀJÁ
(5.9)
226
CONTROLLABILITY AND OBSERVABILITY
of dimensions ÇÈ&ÉJÊÌËÎÍ/ÏÐÑ2È&ÉJÊË-Í2Ò , ÓÔËÕÍ+ÏÖ9Ï , and ×ØËÎÍ2ÒRÖRÏ . Following the
same arguments as in the previous section, we can conclude that the knowledge
of ÇÙ is sufficient to determine ÇȌÉJÊ at any time instant, since from (5.8) we have
ÇȌÉAÊ'ÚÛ#Ü/ÝßތàÞâá&ãHÇȌÉä
Ê
(5.10)
The problem that we are faced with is to find ÇÈ^É ä Ê from the available measurements (5.9). In Section 5.1 we have solved this problem for discrete-time
systems by generating the sequence of measurements at discrete-time instants
å
ÚçæÐEè9ÐêéKÐEë;ë;ë<Ðêì-í4è , i.e. by producing relations given in (5.3). Note that a
time shift in the discrete-time corresponds to a derivative in the continuous-time.
Thus, an analogous technique in the continuous-time domain is obtained by taking
derivatives of the continuous-time measurements (5.9)
Ñ2ȌÉ^ä
Ê'Ú)×îÇDÈ^É&ä
Ê
ï
ï
ÑÈ&É^ä
Ê'Ú)× ÇDÈ&É^ä
Ê'Ú)×PÓ7ÇDÈ^É&ä
Ê
ð
ð
ÑÈ&É^ä
Ê'Ú)× ÇDÈ&É^ä
Ê'Ú)×PÓMñJÇÈ^É&ä
Ê
(5.11)
..
.
Ñ Ý Ï à‘ò6ã È&É ä Ê'Ú)×Ç Ý Ï à!òã È^É ä Ê'Ú)×BÓ Ï à!ò ÇÈ&É ä Ê
Our goal is to generate ì linearly independent algebraic equations in ì unknowns
of the state vector ÇÈŒÉ ä Ê . Equations (5.11) comprise a system of ìó linear
algebraic equations. They can be put in matrix form as
ôõõ
õõ
õö
Ñ2ï ȌÉ&ä
Ê
Ñð È^É&ä
Ê
ÑÈ^É&ä
Ê
..
.
ù
÷Yøø OÏ Ò Ö
øø Ý ã ò
ø
Ú
ôõõ
÷ øø Ý ÏEÒ ã Ö9ÏNú
×
õöõ
øø
× Ó
P
×PÓ7ñ
ÇȌÉ^äRÊ?Ú@û´ÇÈ^É&ä
Ê'ÚüPÈ^É&ä
Ê
..
ù
.
×PÓ Ï à!ò
(5.12)
Ñ Ý Ï !à òã È&É^ä
Ê
where û is the observability matrix already defined in (5.6) and where the
definition of üPÈ&É^ä
Ê is obvious. Thus, the initial condition ÇȌÉ&ä
Ê can be determined
uniquely from (5.12) if and only if the observability matrix has full rank, i.e.
ýþCÿ û©Ú)ì .
As expected, we have obtained the same observability result for both
continuous- and discrete-time systems. The continuous-time observability theorem, dual to Theorem 5.1, can be formulated as follows.
227
CONTROLLABILITY AND OBSERVABILITY
Theorem 5.2 The linear continuous-time system (5.8) with measurements
(5.9) is observable if and only if the observability matrix has full rank.
It is important to notice that adding higher-order derivatives in (5.12) cannot
increase the rank of the observability matrix since by the Cayley–Hamilton
theorem (see Appendix C) for we have
(5.13)
so that the additional equations would be linearly dependent on the previously
defined equations (5.12). The same applies to the discrete-time domain and
the corresponding equations given in (5.4).
There is no need to produce a test example for the observability study of
continuous-time systems since the procedure is basically the same as in the case
of discrete-time systems studied in the previous section. Thus, Examples 5.1
and 5.2 demonstrate the presented procedure in this case also; however, we have
to keep in mind that the corresponding matrices
and describe systems
which operate in different time domains. Fortunately, the algebraic procedures
are exactly the same in both cases.
5.3 Controllability of Discrete Systems
Consider a linear discrete-time invariant control system defined by
!( "! #
$
%'&
$*)
,+ %-
(5.14)
The system controllability is roughly defined as an ability to do whatever we
want with our system, or in more technical terms, the ability to transfer our
/
%1
system from any initial state .+ to any desired final state 0 in a finite time, i.e. for 3254 (it makes no sense to achieve that goal at
4
). Thus, the question to be answered is: can we find a control sequence
( 6+ ( (
/1
*) 7)989898:) "; , such that <
?
Let us start with a simplified problem, namely let us assume that the input
( !
!
< is a scalar, i.e. the input matrix & is a vector denoted by = . Thus, we have
!CB >? @! #
$
A=
<*)
,+ /-
(5.15)
228
CONTROLLABILITY AND OBSERVABILITY
Taking D@EGFH9IJH7KHMLNLNLOH6P in (5.15), we obtain the following set of equations
Q#R I SEUTWV 0
Q R XF SAY[Z#V9\
Q R KX]
S EUTV QR I SAY[Z#V^\
R FXS
R IS]E Q R XF S<Y'TVMZ#VM\ R FXSAY'Z#VM\ R IS
T V_ 0
(5.16)
..
.
QR P/S0E T"` Q R
R
R P fIS
V XF SAY[T b`V a<c Z V \ F S<Yed9d9d^Y[Z V \ W
The last equation in (5.16) can be written in matrix form as
QR PgSgfhT `V QR FXS0EjikZ#V ... TVMZ#V ... ^d d9d ... T J` alc #
V Z V7m
no R
o\ o R P fIS
o \ Pf[KS
op
qsr
\ R I S
\ R FS
t
..
.
r
r
r
r
(5.17)
.
.
.
Note that iuZ V .. T V Z V .. ^d d9d .. T `bV alc Z V m is a square matrix. We call it the controllability matrix and denote it by v . If the controllability matrix v is nonsingular,
equation (5.17) produces the unique solution for the input sequence given by
no R
o\ o R P fIS
o \ PfwKS
op
qr
\ R I S
\ R FS
t
..
.
r
r
r
r
Q R F S.S
Exvayc RzQR PgSgfhT"`V (5.18)
Thus, for any QR PgS{E Q/| , the expression (5.18) determines the input sequence
that transfers the initial state Q/} to the desired state Q | in P steps. It follows
that the controllability condition, in this case, is equivalent to nonsingularity of
the controllability matrix v .
In a general case, when the input ~ R D$S is a vector of dimension  , the
repetition of the same procedure as in (5.15)–(5.17) leads to
QR PgS/f[T"` Q0R FXS]E5iu€ V ... T V € V ... 9d d^d ... 
T `bV a<‚ € V m
V
on R
o~ 
o R P fI?S
o ~ PfwKXS
op
..
.
~ R ?I S
~ R FXS
qr
r
r
r
t
r
(5.19)
229
CONTROLLABILITY AND OBSERVABILITY
The controllability matrix, in the general vector input case, defined by
ƒ„6…"†b‡7ˆ†9‰‹Š5Œuˆ>† .. …†Mˆ† ..
. …"Žb†  ˆ>†:‘
. 9^ ..
.
(5.20)
is of dimension ’”“–•3/’ . The corresponding system of ’ linear algebraic equations in •—’ unknowns for ’U• -dimensional vector components of
˜ „6™‰:‡ ˜ „›š?‰7‡9œ9œ9œ:‡ ˜ „ ’" š‰ , given by
ƒ%ŽžŸ Ž¡7¢g¤£
¤
¤
¤
¤¥
§s¨ sŸ Ž^¡*¢,žª
¨
¨
¨
¨
Š
š‰
˜ „W
’ 
˜ „ ’"[¦ ‰
..
.
˜ „:š‰
˜ „.™‰
©
«0„ ‰ …"Ž† «„.™‰]Š¬«/­
…Ž† «„,™X‰
’ 

«/­
(5.21)
ƒ
will have a solution for any
if and only if the matrix has full rank, i.e.
®*¯?°± ƒ²Š ’ (see Appendix C).
The controllability theorem is as follows.
Theorem 5.3 The linear discrete-time system (5.14) is controllable if and
only if
®*¯?°± ƒ³Š ’
where the controllability matrix
ƒ
(5.22)
is defined by (5.20).
5.4 Controllability of Continuous Systems
Studying the concept of controllability in the continuous-time domain is more
challenging than in the discrete-time domain. At the beginning of this section we
will first apply the same strategy as in Section 5.3 in order to indicate difficulties
that we are faced with in the continuous-time domain. Then, we will show how
to find a control input that will transfer our system from any initial state to any
final state.
A linear continuous-time system with a scalar input is represented by
«µ
´ ŠU…"«W¶·/¸g‡
«0„,¹,º‰Še«A»
(5.23)
230
CONTROLLABILITY AND OBSERVABILITY
Following the discussion and derivations from Section 5.3, we have, for a scalar
input, the following set of equations
½ ¼ ¾À¿ ½µ¾G½WÃ'Ä#Å
w
¿bÁ
Æ ¾À¿ÈÇ ½µ¾  ½Ã[ÂÄ#Å>Ã'Ä Å ¼
½w
Ç
¿bÁ Ç
..
.
½É ÊÌË#¾
(5.24)
Ê
¼ ¿ 3
½ ¾ Â"Êb½ÃÂ"ÊbÍlÎMÄ#Å>Ã'ÊbÍ Ä ÅÏ
à Ð9Ð^Ð9ÃÄ#ÅÉ ÊbÍlÎ7Ë
Ç
Ê
J¿ Á
The last equation in (5.24) can be written as
½ ÉÑÊË.Ò
Â Ê ½ Ò
Á7Ó#Ô
ÁÕÓ
¾Ö
×Ø Å ÚÉ ÊbÍlÎ*Ë
Ø
Ø Å sÉ ÊJÍ Ë
Ø
Ç
ØÙ
ż Ò
Å Ò
..
.
ÁÕÓ
ÕÁ Ó
ÛsÜ
Ü
Ü
Ü
Ü
Ò
7Á Ó
Ò
ÁÕÓ
(5.25)
Ý
Note that (5.25) is valid for any Á—ÞUßàÁ.áâ7Á6ã?ä with Á.ã free but finite. Thus, the
nonsingularity of the controllability matrix Ö implies the existence of the scalar
input function Å Ò Á7Ó and its å Ôæ derivatives, for any ÁèçeÁ6ãçêé .
For a vector input system dual to (5.23), the above discussion produces the
same relation as (5.25) with the controllability matrix Ö given by (5.20) and with
the input vector ë Ò ÁÕÓÏÞUìîí , that is
Ö%ÊXïðòñ Ê
×Ø ÑÉ ÊJÍyÎÕË
Øë
Ø É ÊbÍ Ë
Øë
Ç
ØÙ
¼
ë Ò
..
.
Á7Ó
ë Ò Á7Ó
ÛsÜ ñ ÊXïÎ
Ü í
Ü
Ü
Ü
Ò
ÁÕÓ
Ò
ÕÁ Ó
Ý
¾
½É ÊÌË Ò
ÁÕÓgÔ
Â"Êb½ Ò
ÁÕÓ
¾ ó Ò
ÁÕÓ
(5.26)
It is well known from linear algebra that in order to have a solution of (5.26),
it is sufficient that
ô*õö ÷
ô*õ?ö÷ùø
Öw¾
Ö ... ó Ò
Á7Óbú
(5.27)
231
CONTROLLABILITY AND OBSERVABILITY
Also, a solution of (5.26) exists for any û#üuý*þ —any desired state at ý —if and only if
ÿ
(5.28)
Equations (5.25) and (5.26) establish relationships between the state and
control variables. However, from (5.25) and (5.26) we do not have an explicit
answer about a control function that is transferring the system from any initial
state üuý^þ to any final state üuý^þ . Thus, elegant and simple derivations
for the discrete-time controllability problem cannot be completely extended to
the continuous-time domain. Another approach, which is mathematically more
complex, is required in this case. It will be presented in the remaining part of
this section.
From Section 3.2 we know that the state space equation with the control
input has the following solution
üuýÕþ
# $%&$')(
0üuý þ"!
ü*$þ+*
At the final time ý we have
0üuý?þ
# -
,$-. 012$-.%0.'3(
0ü,ý.þ/!
ül
* þ4+*
5
or
5
6$7
98
: 0ü<; # þ
6% '3(
ü*lþ4+>*
=
Using the Cayley–Hamilton theorem (see Appendix
C), that is
D
?6%@BA
D
D
C DFE ü4*lþH
6G
(5.29)
where ü *$þIKJ ML INOIQPRPFPFI 8 N , are scalar time functions, the previous equation
can beG rewritten as
D
5
5
67
8
: #üzý þ
BA
C DFE H
'
D
# $7
G
ü4*lþ ( 4ü *lþ4+>*
232
CONTROLLABILITY AND OBSERVABILITY
or
tu
wW Y
u
u
u W=ax
u
u w WY
u
b yVe|{byVe4}y
o
Tpoqisr u
... j i ... k&klk ... n
j m
uv W=ax
..
.
SOTVUXW$Y4Z\[^]_SOTU`W=a4ZXb c.deXfhg$i
wW Y
W=ax m
~2








d byze|{byVe4}>y
TVo
byVe|{byVe }y
€
On the left-hand side of this equation all quantities are known, i.e. we have a
constant vector. On the right-hand side the controllability matrix is multiplied
by a vector whose components are functions of the required control input. Thus,
we have a functional equation in the form
tu
?‚"ƒ"„…
mV†
o fˆ‡Xb
j‰
~
b{byzee
uvŒ‹ o
b{byzee
‹
i3e m†>Š4m
..
.
‹
m
TŽo
 Š4m†
o
‰
€
ynb$c d
c e
‰ o
(5.30)
b4{byVe|e
A solution of this equation exists if and only if ‘’l“z” ‡Xb j•‰ i–e6f— , which is the
condition already established in (5.28). In general, it is very hard to solve this
equation. One of the many possible solutions of (5.30) will be given in Section
5.8 in terms of the controllability Grammian. The controllability Grammian is
defined by the following integral
˜
b c d
c e6f
‰ o
š W™
W a›
UœW a T&ž i@isŸ
U` œ2W a T&ž }y
(5.31)
›
The results presented in this section can be summarized in the following
theorem.
Theorem 5.4 The linear continuous-time system is controllable if and only if
the controllability matrix ¡ has full rank, i.e. ¢q£¤¥O¡B¦¨§ .
We have seen that controllability of linear continuous- and discrete-time
systems is given in terms of the controllability matrix (5.20). Examining the
rank of the controllability matrix comprises an algebraic criterion for testing
system controllability. The example below demonstrates this procedure.
233
CONTROLLABILITY AND OBSERVABILITY
Example 5.3: Given the linear continuous-time system
»¼
ª © «®­¯^´°
¬
±
²9³
²1µ
¶
¹
º
²¸·
»¼9À
²s´ ±
ª¾½¿­¯<°
³
µ
²
²¸¶
The controllability matrix for this third-order system is given by
Á
.. Ã
.Ä
«hÂ$Ã
«
­È °
¯
..
.
..
.
..
.
²s´ ±
³
µ
..
. ľŠÃÇÆ
²
²¸¶
²¸·
¹
±l³
²É±l°
µ·
²9¶O¶
..
.
..
.
..
.
»2Ê
Ä
¼
Å
Ã
Since the first three columns are linearly independent we can conclude that
ËÌ?ÍÎ Á « ´
. Hence there is no need to compute Ä Å Ã since it is well known
from linear algebra that the row rank of the given matrix is equal to its column
Á «
´ «ÐÏ
rank. Thus, ËÌ͎Î
implies that the system under consideration is
Ñ
controllable.
5.5 Additional Controllability/Observability Topics
In this section we will present several interesting and important results related to
system controllability and observability.
Invariance Under Nonsingular Transformations
In Section 3.4 we introduced the similarity transformation that transforms
a given system from one set of coordinates to another. Now we will show
that both system controllability and observability are invariant under similarity
transformation.
Consider the vector input form of (5.23) and the similarity transformation
ª Ò «ÓǪ
(5.32)
À
such that
Ò© «
ª
Ä
Ò Ò
ª¾½
Ã
Ò
234
CONTROLLABILITY AND OBSERVABILITY
ÞßÖÙØsÞ
and Ô
ÕáÞ3â
The pair à
where
Õ×ÖÙØÚÕÛØÉÜVÝ
Ô
. Then the following theorem holds.
Օá ÞÉä
Ô
is controllable if and only if the pair ã Ô
Theorem 5.5
is controllable.
This theorem can be proved as follows
å
Õá Þ ä
ã9Ô
Ô
Ö¿æ Þ
Ö
.. Õ Þ ..
. Õéè ÜVÝ Ç
Þ ê
Ô
. Ô Ô . &ç ç&ç .. Ô
Ô
.. ØÚÕÛØ ÜVÝ ØsÞ ..
. Ø:Õ3è ÜVÝ Ø ÜVÝ ØsÞ ê
.
. &ç çlç ..
æ ØsÞ
ÖMØ
.. Õ¾Þ ..
. Õ è ÜVÝ Þ ê ÖØ å Օá|ޖâ
à
.
. &ç çlç ..
æ Þ
Ø
Since
is a nonsingular matrix (it cannot change the rank of the product
we get
ë?
ì íî
ëq
ì íî
å
å
Õá s
Þ äÖ
ãÔ
Ô
à
Ø å
),
Õá|Þ@â
which proves the theorem and establishes controllability invariance under a
similarity transformation.
A similar theorem is valid for observability. The similarity transformation
(5.32) applied to (5.8) and (5.9) produces
ï
ð
ñ
where
Ö
Ô
Õ
Ö
ð
Ô Ô
ò ð
Ô Ô
òóÖMòôØ
Ô
ÜVÝ
Then, we have the following theorem
Օá ò ä
Ô
Õáò@â
is observable if and only if the pair ã1Ô
Theorem 5.6 The pair à
is observable.
The proof of this theorem is as follows
õ
Õ á ò ä
ãöÔ
Ô
Ö®÷ø
ò
Ô
ò Õ
Ô Ô
ò Õ¾ú
Ô Ô
øù
ø
ø
..
.
ò Õ è ÜVÝ
Ô Ô
û2ü
ò@Ø ÜVÝ
û2ü
ò
ü
ò@Ø ÜVÝ ØÚÕÛØ V
Ü Ý
ú
ò@Ø ÜVÝ ØÚÕ
Ø V
Ü Ý
ü
òôÕ
ü
ý
ü
Ö®÷ø
ø
øù
ø
ò@Ø
ÜVÝ ØÚÕ
..
.
è ÜVÝ Ø
ÜVÝ
ü
ý
ü
û2ü
ü
ü
òôÕ
Öþ÷ø
ø
øù
ø
òôÕ
ü
ú
..
.
è ÜVÝ
ý
ؖ܎Ý
235
CONTROLLABILITY AND OBSERVABILITY
ÿ
ÿ
that is,
The nonsingularity of
ÿ
implies
ÿ
which proves the stated observability invariance.
Note that Theorems 5.5 and 5.6 are applicable to both continuous- and
discrete-time linear systems.
Frequency Domain Controllability and Observability Test
Controllability and observability have been introduced in the state space
domain as pure time domain concepts. It is interesting to point out that in the
frequency domain there exists a very powerful and simple theorem that gives a
single condition for both the controllability and the observability of a system.
It is given below.
Let
be the transfer function of a single-input single-output system
#"
! #"
! $
&%
"
! '")(+*
, -).
Note that
is defined by a ratio of two polynomials containing the corresponding system poles and zeros. The following controllability–observability
theorem is given without a proof.
Theorem 5.7 If there are no zero-pole cancellations in the transfer function
of a single-input single-output system, then the system is both controllable and
observable. If the zero-pole cancellation occurs in
, then the system is either
uncontrollable or unobservable or both uncontrollable and unobservable.
A similar theorem can be formulated for discrete linear time invariant
systems.
Example 5.4: Consider a linear continuous-time dynamic system represented
by its transfer function
"
! '"0/21
"6/51
'"+/&3 #"0/24 #"+/51 "87+/29:");</=3:38"+/>9
$
!
#"
Theorem 5.7 indicates that any state space model for this system is either
uncontrollable or/and unobservable. To get the complete answer we have to go
236
CONTROLLABILITY AND OBSERVABILITY
to a state space form and examine the controllability and observability matrices.
One of the possible many state space forms of
is as follows
FG-IKH J OPRQSFGTU T VEV
V W
IKH L
W V
INH M
Q_^
] W V
?A@CBED
TU PX
O GF -I J OPYZFGV OP\[
W IKL
W
W INM
W
FG I J OP
`a I L
IM
It is easy to show that the controllability and observability matrices are given by
b
Since
and
QcFGV T U d:e OPgf i
h QSFGjW
W V TU
V
W W V
T`
kmln b Q VQ o Wqp r stu b Q
h x
kmln hiQ Wqp rswtu y
`
this system is controllable, but unobservable.
Note that, due to a zero-pole cancellation at
function
is reducible to
?A@#BwD
Q
Q
` PO
W
T V:V TU
Q&v
`
B
`
V
Q&v
Q
T` , the system transfer
Y V Y Q Y V Y
?A@BD ?z)@'BD #@ B Vv|D{@'Q B d D B L ` B d
d has the corresponding state space
so that the equivalent system of order
} I H J z Q } Td T` } I J z Y } V [
form
V W ~ I L zK~ W ~
I H L zN~
Qq^
}
] W V a II LJ z
z-~
For this reduced-order system we have
Q } V Td fh€Q } W V
b W V
V W~
~
237
CONTROLLABILITY AND OBSERVABILITY
and therefore the system is both controllable and observable.
Interestingly enough, the last two mathematical models of dynamic systems
of order
and
represent exactly the same physical system. Apparently,
the second one (
) is preferred since it can be realized with only two
integrators.
,‚&ƒ
‚=„
…‚†„
‡
It can be concluded from Example 5.4 that Theorem 5.7 gives an answer
to the problem of dynamic system reducibility. It follows that a single-input
single-output dynamic system is irreducible if and only if it is both controllable
and observable. Such a system realization is called the minimal realization. If
the system is either uncontrollable and/or unobservable it can be represented
by a system whose order has been reduced by removing uncontrollable and/or
unobservable modes. It can be seen from Example 5.4 that the reduced system
is both controllable and observable, and hence it cannot be further
with
.
reduced. This is also obvious from the transfer function
Theorem 5.7 can be generalized to multi-input multi-output systems, where
it plays very important role in the procedure of testing whether or not a given
system is in the minimal realization form. The procedure requires the notion of
the characteristic polynomial for proper rational matrices which is beyond the
scope of this book. Interested readers may find all details and definitions in
Chen (1984).
It is important to point out that the similarity transformation does not change
the transfer function as was shown in Section 3.4.
Controllability and Observability of Special Forms
|‚€„
ˆ‰8Š#‹EŒ
In some cases, it is easy to draw conclusions about system controllability
and/or observability by examining directly the state space equations. In those
cases there is no need to find the corresponding controllability and observability
matrices and check their ranks.
Consider the phase variable canonical form with
Ž  ‚& Ž‘>’“
” ‚–• Ž
where
238
CONTROLLABILITY AND OBSERVABILITY
ž ©«ª
8 ) ž ©«ªª
8 ) ž ªª
ž ªªª
—™˜›šœœ ... ... ... . . . ...
˜ œš .
›
œœ ž ž ž 8 ) Ÿ ¬­ ® œœœ ž .. ¬
Ÿ
¡¢:£ ¡¢¥¤ ¡¢¥¦ 8 ) ¡ ¢:§:¨-¤
¯°˜_± Ÿ ž ž ²)²8²³žµ´
ž
ž
ž
Ÿ
ž
Ÿ
¶ ¤8·¹¸º ˜ ½ ˜ ¿
˜
½
¦
¤
¼
·
{
¸
º
¼
·
{
¸
º
¶
¶
¶N¾ ·¸»º ¶ ¦ ·¸»º ¶ ¤ ·¼¸{º
:
§
¨
¤
Â
˜
¶ §-·¹¸»º Á¶ ¤À ·¹¸»º
à ·¸»º
¶ ½ :§ ¨-¤8·¸»º ¶ §· à ·¸{º»º
¶ ¦8·¸»º
This form is both controllable and observable due to an elegant chain connection
of the state variables. The variable
is directly measured, so that
is known from
. Also,
, and so on,
. Thus, this form is observable. The controllability follows
is
from the fact that all state variables are affected by the control input, i.e.
affected directly by
and then
by
and so on. The control
input is able to indirectly move all state variables into the desired positions so
that the system is controllable. This can be formally verified by forming the
corresponding controllability matrix and checking its rank. This is left as an
exercise for students (see Problem 5.13).
Another example is the modal canonical form. Assuming that all eigenvalues
of the system matrix are distinct, we have
¶§
˜ Å ÇÄ Æ>ÈÊÉ
Ľ R
Ë ˜–Ì Ä
where
Í ¤ ž ) 8 ž ©«ªª
Î ¤ ©«ªª
ÅX˜ œš ž . Í .¦ ). 8 ž .
˜ œš Î .¦
È
œ ž .. ž .. ). 8. Í .. ¬ ­
œ .. ¬
Χ
§
̳˜_±ÐÏ ¤ Ï ¦ ) 8 Ï § ´
We are apparently faced with Ñ completely decoupled first-order systems. Ob˜ Ÿ ²)²8² Ñ must be different from zero,
viously, for controllability all ÎmÒ
­-Ó controlled
­ ­ by­ the input É ·¼¸{º . Similarly,
so that each state variable can be
Ï Ò+˜ Ô ž ˜ Ÿ ²)²»² Ñ ensures observability
since, due to the state decomposition,
­
Ó
­
­
­
each system must be observed independently.
CONTROLLABILITY AND OBSERVABILITY
239
The Role of Observability in Analog Computer Simulation
In addition to applications in control system theory and practice, the concept
of observability is useful for analog computer simulation. Consider the problem
of solving an th-order differential equation given by
Õ
Ý Ý
Ýìë Ý
Ø
Ö׫Ø:ÙÛÚ Ü ÝßÞáàµâ Ø:ã NÖ ×äØEã Ùæå Ü ÝèÞKç é$ê ã à × ç ã Ù
ç
N
Ö
#
í
m
î
ï
á
ð
m
Ö
ñ
#
í
ò
î
{
ï
ô
ð
õ
ó
è
ó
è
ó
ð
Ö
×ÐØ:ã Ù í#îòï . This system can be
with known initial conditions for
à
solved by an analog computer by using Õ integrators. The outputs of these Õ
integrators represent the state variables ö ð ö-÷ ð)óõóèóèð ö Ø ð so that this system has the
state space form
ë
ñø å&ù ø Ú>ú ð ø í#îòïûåRüKýþ¥ýÿ 6ý
Ö
å ø
However, the initial condition for ø í#îòï isà not given. In other words, the
initial conditions for the considered system of Õ integrators are unknown. They
can be determined from ÖíîòïðáÖmñ í'îmïð)óèóõóèðÖ ×äØEã Ù í#îòï by following the observability
derivations performed in Section 5.2, namely
ë
ÖNí'îmïûå ø í#îmï
ë
NñÖ í'îmïû
å +ñø í'îmïûå:ù ø í#îòïÛÚ:ú í#îòë ï
ÖN í'îmïû
å ø í'îmïû
å :ù ÷ ø í#îòïÛÚ:ù‘ú íîòïáÚ:úñ í'îmï
à
à
à
..
ë
.
Ö × Ø:ã Ù í#îmïµ
å ø × Ø:ã Ù í'îmï$åë :ù Ø:ã ø í'îmïÛÚ:ù ë Ø:ã ÷ ú í#îòï ë
Ú Eù ØEã )úñ í#îòïÛÚÚEù‘ú ×äØEã
Ù'í'îmïÛÚ:ú × Ø:ã ÷ Ù'í'îmï
This system can be written in matrix form as follows
ë î
Öíîòï ñÖ íà îòï å ø íîòïÛÚ
í#îòï ..
..
ë
.
.
Ö ×«ØEã Ù íîmï
× Ø:ã ÷ Ù í#îòï
ë
(5.33)
à
ë is the
ë observability
matrix and is a known matrix. Since
í#îòï{ðÛñ í#îòï{ðôóõóèóèð ×ÐØ:ã Ù í'îmï are known,
it follows that a unique solution for ø íîòï
where 240
CONTROLLABILITY AND OBSERVABILITY
exists if and only if the observability matrix, which is square in this case, is
invertible, i.e. the pair ( ) is observable.
Example 5.5: Consider a system represented by the differential equation
!#"
*"
!.
"-,
.
"0/21!34,5
"0/81!39,;:
.</=$34,>@?BADC $FEG1
%$ '&)( +$ &(
+$ &
76
Its state space form is given by
H6 ,
1
.',KJ
H &I
L (
"-,
,OP:
H
:
L N
( M
:<Q H
H &
J
1
:
.
M
The initial condition for the state space variables is obtained from (5.33) as
/813
1
5
1
, J
J J
H /D1!3R, JTS /813 L J
.</D1!3
: L
:
S6
% M
I
M
M
M
M
leading to
J
:
:
L (
LVU M
5
H /81!3R,KJ 1
M
W
/21!3
L ^
H /2134,XJZY\[ /213 ,]J _
`
Y
M
M
This means that if analog computer simulation is used to solve the above secondorder differential equation, the initial conditions for integrators should be set to
La^ and ` .
b
Stabilizability and Detectability
So far we have defined and studied observability and controllability of the
complete state vector. We have seen that the system is controllable (observable)
if all components of the state vector are controllable (observable). The natural
question to be asked is: do we really need to control and observe all state
variables? In some applications, it is sufficient to take care only of the unstable
components of the state vector. This leads to the definition of stabilizability and
detectability.
Definition 5.1 A linear system (continuous or discrete) is stabilizable if all
unstable modes are controllable.
CONTROLLABILITY AND OBSERVABILITY
241
Definition 5.2 A linear system (continuous or discrete) is detectable if all
unstable modes are observable.
The concepts of stabilizability and detectability play very important roles in
optimal control theory, and hence are studied in detail in advanced control theory
courses. For the purpose of this course, it is enough to know their meanings.
5.6 Observer (Estimator) Design1
Sometimes all state space variables are not available for measurements, or it is not
practical to measure all of them, or it is too expensive to measure all state space
variables. In order to be able to apply the state feedback control to a system,
all of its state space variables must be available at all times. Also, in some
control system applications, one is interested in having information about system
state space variables at any time instant. Thus, one is faced with the problem
of estimating system state space variables. This can be done by constructing
another dynamical system called the observer or estimator, connected to the
system under consideration, whose role is to produce good estimates of the state
space variables of the original system.
The theory of observers started with the work of Luenberger (1964, 1966,
1971) so that observers are very often called Luenberger observers. According
to Luenberger, any system driven by the output of the given system can serve
as an observer for that system. Two main techniques are available for observer
design. The first one is used for the full-order observer design and produces
an observer that has the same dimension as the original system. The second
technique exploits the knowledge of some state space variables available through
the output algebraic equation (system measurements) so that a reduced-order
dynamic system (observer) is constructed only for estimating state space variables
that are not directly obtainable from the system measurements.
5.6.1 Full-Order Observer Design
Consider a linear time invariant continuous system
dfc eZgih9jkldfemghonprqse2ghut
dfe2g2vh9jdxwajGyzB{!z|}_z
~ emgh4j€dse2gh
1
This section may be skipped without loss of continuity.
(5.34)
242
CONTROLLABILITY AND OBSERVABILITY
where ‚„ƒ†…ˆ‡ , ‰„ƒŠ…F‹ , ŒƒŠ…Ž with constant matrices ‘’l‘#“ having
appropriate dimensions. Since from the system (5.34) only the output variables,
Œ9”=•u– , are available at all times, we may construct another artificial dynamic
system of order — (built, for example, of capacitors and resistors) having the
same matrices ˜‘u’l‘#“
š™
š
š
š
‚9”=•–9›œ ‚s”=•–x’ž‰N”=•–i‘
‚f”2•2Ÿ–9› ‚N (5.35)
š
š
Œ9”2•–4›“ ‚s”=•–
š
and compare the outputs Œ”m•– and Œf”2•– . Of course these two outputs will be
different since in the first case the system’s initial condition is unknown, and in
the second case it has been chosen arbitrarily. The difference between these two
outputs will generate an error signal
š
š
Œ”m•–R¡ Œ”2•–f›¢“£‚s”=•–R¡¤“ ‚9=” •–4›“¥B”=•–
(5.36)
which can be used as the feedback signal to the artificial system such that the
š
estimation (observation) error ¥\”2•–4›G‚9”=•–¡ ‚9”m•– is reduced as much as possible.
This can be physically realized by proposing the system-observer structure as
given in Figure 5.1.
u
F
System
B
Ce
K
¦
y=Cx
+
-
Observer
§x x
Figure 5.1: System-observer structure
y=Cx
CONTROLLABILITY AND OBSERVABILITY
243
In this structure ¨ represents the observer gain and has to be chosen such
that the observation error is minimized. The observer alone from Figure 5.1 is
given by
ª © ¬Z­i®9¯° «f
ª ¬2­®o±²r³s¬2­®x±
ª ¬2­®o±²r³s¬=­®o±
«f
¶´ ¬=­®®R¯° «f
¨ ¬2´¬m­®Rµ4
¨¸·¹ ¬=­® (5.37)
From (5.34) and (5.37) it is easy to derive an expression for dynamics of the
estimation (observation) error as
©
¹ 2¬ ­®f¯¢¬8°ºµ ¨¸· ® ¹ ¬2­®
(5.38)
If the observer gain ¨ is chosen such that the feedback matrix °ºµ ¨¸· is
asymptotically stable, then the estimation error ¹ ¬=­® will decay to zero for any
initial condition ¹ ¬=­8»® . This can be achieved if the pair ¬8°¼ · ® is observable.
More precisely, by taking the transpose of the estimation error feedback matrix,
i.e. °¾½¤µ · ½ ¨ ½ , we see that if the pair ¿ °l½R¼ · ½ÁÀ is controllable, then we
can do whatever we want with the system, and thus we can locate its poles in
arbitrarily asymptotically stable positions. Note that controllability of the pair
¿ °l½R¼ · ½ À is equal to observability of the pair ¬8°˜¼ · ® , see expressions for the
observability and controllability matrices.
In practice the observer poles should be chosen to be about ten times faster
than the system poles. This can be achieved by setting the minimal real part of
observer eigenvalues to be ten times bigger than the maximal real part of system
eigenvalues, that is ÃÅÄÆ+Ç*ÈFÉËÊÌ ÍDÎmÏ2ÐÒÑDÓuÐmÑ£Ô¢ÕÖ\ Ã×ÄÆ+ÇȏØÙ!Ì Ï=ÚuÏ=ÛÜÐ È
Theoretically, the observer can be made arbitrarily fast by pushing its eigenvalues
far to the left in the complex plane, but very fast observers generate noise in
the system. A procedure suggesting an efficient choice of the observer initial
condition is discussed in Johnson (1988).
It is important to point out that the system-observer structure preserves the
closed-loop system poles that would have been obtained if the linear perfect
state feedback control had been used. The system (5.34) under the perfect state
feedback control, i.e. ³s¬2­®ˆ¯Ýµ_ވ«9¬=­® has the closed-loop form as
«9© ¬=­®4¯¬2°ºµ²ßÞa®=«9¬2­®
(5.39)
244
CONTROLLABILITY AND OBSERVABILITY
so that the eigenvalues of the matrix àâáã-ä are the closed-loop system poles
under perfect state feedback. In the case of the system-observer structure, as
given in Figure 5.1, we see that the actual control applied to both the system
and the observer is given by
åsæ2çèé
áaäžës
ê æ2çèfé
áVä ësæ2çèíì äíî æ2çè
so that from (5.34) and (5.38) we have
ï%ë ð
ï
ïë
º
à
á
ã
ä
ã
ä
é
ð
ò
îFñ
àºá)ó¸ô)ñ îFñ
(5.40)
(5.41)
Since the state matrix of this system is upper block triangular, its eigenvalues are
equal to the eigenvalues of matrices à¤á£ãßä and à¤áó¸ô . A very simple relation
among ë9õ î õ and ë ê can be written from the definition of the estimation error as
ïë
ïm÷
ïë
ïë
ò
é
é
ø
÷
÷
(5.42)
ëê ñ
îöñ
á ñ ëê ñ
Note that the matrix ø is nonsingular. In order to go from ë î -coordinates to
ë ë ê -coordinates we have to use the similarity transformation defined in (5.42),
which by the main property of the similarity transformation indicates that the
same eigenvalues, i.e. ù æ àúá)ãßä è and ù æ àúá¤ó¸ô è , are obtained in the ë ë ê coordinates.
This important observation that the system-observer configuration has closedloop poles separated into the original system closed-loop poles under perfect state
feedback and the actual observer closed-loop poles is known as the separation
principle.
5.6.2 Reduced-Order Observer (Estimator)
In this section we show how to construct an observer of reduced dimension
by exploiting knowledge of the output measurement equation. Assume that the
output matrix ô has rank û , which means that the output equation represents û
linearly independent algebraic equations. Thus, equation
ü æmçè4é
ô ësæ2çè
(5.43)
produces û equations for ý unknowns of the state space vector ëfæmçè . Our goal
is to construct an observer of order ý'¤
á û for estimation of the remaining ý'áû
state space variables.
245
CONTROLLABILITY AND OBSERVABILITY
The reduced-order observer design will be presented according to the results
of Cumming (1969) and Gopinath (1968, 1971). The procedure for obtaining
this observer is not unique, which is obvious from the next step. Assume that
a matrix þ£ÿ exists such that
þ
(5.44)
þ ÿ
and introduce a vector as
þ-ÿ
(5.45)
From equations (5.43) and (5.45) we have
ÿ !
þ
þ ÿ (5.46)
Since the vector is unknown, we will construct an observer to estimate it.
Introduce the notation
ÿ
þ
ÿ
(5.47)
"%$
$'&)(
þ ÿ #
so that from (5.46) we get
*$
ÿ
!
,+
$ &
(5.48)
An observer for can be constructed by finding first a differential equation
for - from (5.45), that is
.
þ-ÿ .
þ-ÿ/01+
þ ÿ243
þ ÿ/
$ &
5+þ ÿ6/
$
ÿ
!
+Gþ-ÿ6273
(5.49)
!
Note that from (5.49) we are not able to construct an observer for since does not contain explicit information about the vector , but if we differentiate
the output variable we get from (5.34) and (5.48)
!.
i.e.
!.
þ .
þ8/4
+Gþ9273
þ9/
carries information about .
$ &
4+
þ8/
$
ÿ
!
+þ9273
(5.50)
246
CONTROLLABILITY AND OBSERVABILITY
An observer for :';<= , according to the observer structure given in (5.37), is
obtained from the last two equations as
?>
?
>
?>
:A@B9CD1EGF :IJ
H B8C6DKEGCMLIHJB8C6N7O1HQPICSR LUT W
L V
(5.51)
where P C is the observer gain. If in equation (5.50) we replace :;<= by its
estimate, we will have
?>
?
LA@B8DKEGF :KHXB9D1EYCL4HZB8N7O
(5.52)
so that
?>
?
>
?
:U@[B8CD1EGF :IJ
H B8C6DKEGCLKHJB9C6N7O4HQPIC\; L5J
T B9D1EGF :5]
T B8DKEGCLKTJB9N7O=
(5.53)
>
Since it is impractical and undesirable to differentiate L';<= in order to get L;<=
(this operation introduces noise in practice), we take the change of variables
_^ @
?
:ITQPICL
^
This leads to an observer for _ ;<= of the form
`
_ ^ ;<=@aD1b _ ^ ;<=cHQN9b\Od;<=,HJPIbL;<=
(5.54)
(5.55)
where
D b @[B8CD1EYFTQPIC\B9D1EGFfe
N b @B8CNgTQPICfB8N
P b @B9C6D1EGFPICHXB9CD1EGChTQPIC\B8DKEGCTQPIC\B9D1EGFPIC
(5.56)
It is left as an exercise to students (see Problem 5.18) to derive (5.55) and (5.56).
The estimates of the original system state space variables are now obtained from
(5.48) and (5.53) as
^
^
i ? ;-<=h@*EGCL';<=cHQEGF :
; <=@aEGF _ ; <=cHX;EYChHQEGF6PIC\=L
(5.57)
The obtained system-reduced-observer structure is presented in Figure 5.2.
There are other ways of constructing the system observers (Luenberger, 1971;
Chen, 1984). The reader particularly interested in observers is referred to a
specialized book on observers for linear systems (O’Reilly, 1983).
247
CONTROLLABILITY AND OBSERVABILITY
ju
y
System
B
k
Kq
m
F
Bq
q
Reduced
observer
L2
L1+L2K1
+
l+
x
x
Figure 5.2: System-reduced-observer structure
5.7 MATLAB Case Study: F-8 Aircraft
In the case of high-order systems npoUqXrts , obtaining the controllability and observability matrices is computationally very involved. The MATLAB package for
computer-aided control system design and its CONTROL toolbox help to overcome this problem. Moreover, use of MATLAB enables a deeper understanding
of controllability and observability concepts. Consider the following fourth-order
model of an F-8 aircraft studied in Teneketzis and Sandell (1977), Khalil and
Gajić (1984), Gajić and Shen (1993). The aircraft dynamics in continuous-time
is described by the following matrices
‘%’
’
r‚tƒ
r†… …
r
uwvyxz
…
…
z{}|~€~
~„~
|

|G‡‰ˆ ~†~
~€~†~„~†~ “
… …
…
~Š%~†~„~ ~„~
~Š%~
„ f‡
~€~†~„~†~
‚ƒ
|‹~€~†~†~Š 
~Š%~
|Œ† 6‡
†€~†~„~†~
~„~
~Š%~
|ŽŠ%~Š~
|‹~%ˆ„ˆ†Ž„ˆ
”•v#– ~Š%~†~„~
r†r
r
r
‚ƒ†ƒ›šœ
|~Š—‡
~Š˜ Ž™‡
gvŸž
~
|‹~š

~
~
~
~
Ž ‡

~
|‹~Š˜
248
CONTROLLABILITY AND OBSERVABILITY
By using the MATLAB function ctrb (for calculation of the controllability
matrix ¡ ) and obsv (for calculation of the observability matrix ¢ ), it can be
verified that this system is both controllable and observable, namely
£¤™¥¦ ¡W§¨„§ª©*«¬
£¤™¥¦ ¢®­¯¨„§‹©X«
By using the MATLAB function det (to calculate a matrix determinant), we get
°t±6²
¡³©#´‹µt¶%·†µ†¸†¹
Since ºt»6¼½¡ is far from zero it seems that this system is well controllable (the
controllability margin is big).
If we discretize the continuous-time matrices ¾ and ¿ by using the sampling
period ÀÂÁg©Ã¹Š¶˜Ä , we get a somewhat surprising result. Namely
°t±f²
ÐÕÎ
¡Æž1Ǭ¿9ǐÈfÉ ÊWËÍÌSÎfÏ Ð©[´«¶ ц҄·ÄªÓQÄ\¹†Ô
Thus, this discrete system is almost uncontrollable. Theoretically, it is still
controllable but we need an enormous amount of energy in order to control it.
For example, let the initial condition be ÖŹÈY©Ø×šÄ Ä
Ä
ÄWÙ and let the final
state ÖÅ«ŠÈ be the coordinate origin. Then, by (5.18), the control sequence that
solves the problem of transferring the system from ÖŹtÈ to ÖÅ«ŠÈY©ÃÚ , obtained
ÛÜ
by using MATLAB is ÛÜ
Ü
Ü
à%á
à%á
ÅÄ™È á
Ҋ¶˜Äµ„·†ß á
ÅpßtÈ
´‹¸Š¶%含tµ«
ÝpÞ ÅpÒÈ â ©[Đ¹†ãäÓ Ý ¸Š¶%¸†¸™«†« â
Þ Å«ŠÈ
Ҋ¶—«SÄ\µ†Ò
Þ
Apparently, this result is Þ unacceptable and this discrete system is practically
uncontrollable.
Note that the eigenvalues of the continuous-time controllability Grammian (5.31), obtained by using the MATLAB function gram, have values
åt¶—«‰·ŒÓQĐ¹ § ¬h¹Š¶—«‰ßŠ¬Æ¹¶€¹†Òtåt¬Æ¹¶%¹„¹†µ„· . The eigenvalues of the controllability Grammian are the best indicators of the controllability measure. Since two of them
are very close to zero, the original system is very badly conditioned from the
°±6²
controllability point of view even though
¡Å¾U¬¿7È is far from zero. The
interested reader can find more about controllability and observability measures
in a very comprehensive paper by Muller and Weber (1972).
249
CONTROLLABILITY AND OBSERVABILITY
5.8 MATLAB Laboratory Experiment
Part 1. The controllability staircase form of the system
æ è*é0ç1êìë7í
çU
î èïŒç
clearly distinguishes controllable and uncontrollable parts of a control system. It
can be obtained by the similarity transformation, and is defined by
ð çcæ ñ
ð é1ñ é0ôõ ð çcñ
ð ë9ñ
è
ê
ö
ö ó í
çWæ ò ñ„ó
é0ò ñ„ó ç÷ò ñ„ó
ð ç ñ
(5.58)
î ègø€ï ñ ïùò ñÕú
ç÷ò ñ ó
where ç ñ are controllable modes, and çWò ñ are uncontrollable modes. Apparently,
in this structure the input í cannot influence the state variables ç ò ñ ; hence these
are uncontrollable. Similarly, one can define the observability staircase form as
ð çWæ û
ð éKû
ð çdû
ð ë9û
ö
è
ê
í
çWæ ò û ó
é õô é0ò û ó ç÷ò û ó
ëŒò û ó
ð ç û
(5.59)
î è#øüï û ö ú
çWò ûó
with ç û observable and ç÷ò û unobservable. Due to the fact that only ç û appears
in the output and that ç û and ç÷ò û are not coupled through the state equations,
the state variables ç÷ò û cannot be observed.
Use the MATLAB functions ctrbf (controllable staircase form) and obsvf
(observable staircase form) to get the corresponding forms for the system
þÿ þÿ ÿ
ÿ
ÿ ÿ
ÿ
ÿ ÿ ÿ ÿ
ÿ
éýè
ë•è
ï•è#ø Identify the corresponding similarity transformation.
ú
250
CONTROLLABILITY AND OBSERVABILITY
Part 2. Derive analytically that the transfer function of (5.58) is given in
terms of the controllable parts, i.e. it is equal to
"!#$&%('*)+,-&./012
"!31%4'*)
(5.60)
Clarify your answer by using the MATLAB function for the transfer function
ss2zp, i.e. show that both transfer functions have the same gains, poles, and
zeros (subject to zero-pole cancellation).
Do the same for the observable staircase form, i.e. show that
6578/599:;!#5* %<' )=5:-8
(5.61)
and justify this identity by using the MATLAB function ss2zp.
Part 3. Examine the controllability and observability of the power system
composed of two interconnected areas considered in Geromel and Peres (1985)
and Shen and Gajić (1990)
?@ B
BDCEFE
@ B
B
@
@
@ B
C I
@ B
@
B
I
I
I
B
B
B
O
B
B
L
B
B
B
JI
B
B
B
B
B
B
B
B
B
H
B
C I
B
B
H I
B
B
B
B
B
B
B
B
B
B
B
JI
B
B
B
B<CBFE
B
I
O
B
B
B<CYX
O
O
O
B
H
B
O
H I
B
@
B
O
B
A B
?@
B
NO
B
B
B
B
B
B
B
B
B
B
B
)QSR
B
B
C I
B
C I
B
C I
E<C M
B
B
C I
EGCE
B
B
JI
B
@
B
L
B
@ B
@
B
BGCKBFE
B
@
!>
B
B
H
JI
@
B
EGCKM
B
B
B
B
B
B
B
B
B
B
B
B
B
B
B<CYX
B
B
B
B
B
O
C I
P
H I
N O
B
H
O
B
B
B
I
O
H IUTGV
B
B
C I
B
O
O
W
A
B
H
I
B
H
P
B
Part 4. Follow the steps used in Section 5.7, but this time for the F-15
aircraft, whose state space model was presented in Example 1.4. Consider both
the subsonic and supersonic flight conditions. Comment on the results obtained.
251
CONTROLLABILITY AND OBSERVABILITY
Part 5. The controllability Grammian is defined in (5.31) as
h
Z\[]_^F`a]&b7c/d
l
egf
m
l+uGm
ejionGp2qrrts
eji6k
ejionvp2q_wyx
k
(a) Show analytically that the control input given by
z
l/uGm
[]$c{d}|
r~s
b
Z
ejionGeq
‚[_] ^
[€] ^
`o] b cGƒ‚[_] ^
c„|
l…m
‚{[] b c†
eji1nGe f q
n
c
(5.62)
‚[_] b cd‡‚„ˆ
into any desired finalk state
. Note that
will drive any initial k state
under the controllability assumption many control inputs can be found to transfer
the system from the initial to the final state. The expression given in (5.62) is
also known as the minimum energy control (Klamka, 1991) since in addition to
‚[€] ^ c
‚{[] b c/d‡‚ ˆ
driving the system from
to
, it also minimizes an integral of the
€
[
]ac z [€]ac
[] ^ `a] b c
z
square of the input (energy),
, in the time interval
.
s
(b) Using the MATLAB function gram, find the controllability Grammian
] ^
dŠ‰
] b
d
‹
for the system defined in Part 4 for
and
. One of several
known controllability tests states that the system is controllable if and only if its
controllability Grammian is positive definite (Chen, 1984; Klamka, 1991). Verify
whether or not the controllability Grammian for this problem is positive definite.
(c) Find the control input Œ„_Ž that drives the system defined in Part 4
to the final state ‘{›8”“
from the initial condition ‘&’v”“–•K’ ’
’
’
’˜—š™
•œ›
›
›
›
›—š™ .
Part 6. By duality to the controllability Grammian, the observability
Grammian is defined as
¤
ž
€Ž1ŸF aŽ¡9/“
¢€£
¦8§4¨©<ª
§G«2¬1­
¢j¥
¢j¥
™
­¦8§G¨…ª
§v«2¬®D¯
(5.63)
¢j¥
Note that the observability Grammian is in general a positive semidefinite matrix.
It is known in the literature on observability that if and only if the observability
Grammian is positive definite, the system is observable (Chen, 1984). Check the
observability of the system given in Part 3 by using the observability Grammian
test.
252
CONTROLLABILITY AND OBSERVABILITY
5.9 References
Chen, C., Introduction to Linear System Theory, Holt, Rinehart and Winson,
New York, 1984.
Chow, J. and P. Kokotović, “A decomposition of near-optimum regulators for
systems with slow and fast modes,” IEEE Transactions on Automatic Control,
vol. AC-21, 701–705, 1976.
Cumming, S., “Design of observers of reduced dynamics,” Electronic Letters,
vol. 5, 213–214, 1969.
Gajić, Z. and X. Shen, Parallel Algorithms for Optimal Control of Large Scale
Linear Systems, Springer-Verlag, London, 1993.
Geromel, J. and P. Peres, “Decentralized load-frequency control,” IEE Proc.,
Part D, vol. 132, 225–230, 1985.
Gopinath, B., On the Identification and Control of Linear Systems, Ph.D. Dissertation, Stanford University, 1968.
Gopinath, B., “On the control of linear multiple input–output systems,” Bell
Technical Journal, vol. 50, 1063–1081, 1971.
Johnson, C., “Optimal initial conditions for full-order observers,” International
Journal of Control, vol. 48, 857–864, 1988.
Kalman, R., “Contributions to the theory of optimal control,” Boletin Sociedad
Matematica Mexicana, vol. 5, 102–119, 1960.
Khalil, H. and Z. Gajić, “Near optimum regulators for stochastic linear singularly
perturbed systems,” IEEE Transactions on Automatic Control, vol. AC-29,
531–541, 1984.
Klamka, J., Controllability of Dynamical Systems, Kluwer, Warszawa, 1991.
Longhi, S. and R. Zulli, “A robust pole assignment algorithm,” IEEE Transactions
on Automatic Control, vol. AC-40, 890–894, 1995.
Luenberger, D., “Observing the state of a linear system,” IEEE Transactions on
Military Electronics, vol. 8, 74–80, 1964.
Luenberger, D., “Observers for multivariable systems,” IEEE Transactions on
Automatic Control, vol. AC-11, 190–197, 1966.
Luenberger, D., “An introduction to observers,” IEEE Transactions on Automatic
Control, vol. AC-16, 596–602, 1971.
253
CONTROLLABILITY AND OBSERVABILITY
Mahmoud, M., “Order reduction and control of discrete systems,” IEE Proc.,
Part D, vol. 129, 129–135, 1982.
Muller, P. and H. Weber, “Analysis and optimization of certain qualities of
controllability and observability of linear dynamical systems,” Automatica, vol.
8, 237–246, 1972.
O’Reilly, J., Observers for Linear Systems, Academic Press, New York, 1983.
Petkov, P., N. Christov, and M. Konstantinov, “A computational algorithm for
pole assignment of linear multiinput systems,” IEEE Transactions on Automatic
Control, vol. AC-31, 1004–1047, 1986.
Teneketzis, D. and N. Sandell, “Linear regulator design for stochastic systems
by multiple time-scale method,” IEEE Transactions on Automatic Control, vol.
AC-22, 615–621, 1977.
Shen, X. and Z. Gajić, “Near optimum steady state regulators for stochastic linear
weakly coupled systems,” Automatica, vol. 26, 919-923, 1990.
5.10 Problems
5.1
Test the controllability and observability of the following systems
°²±´³Jµ
¶…·
±S³1µ
±ÂÁ
¸
¶tµ
¸º¹¼»¾½
ÈÉ
·
¶tµ
¸
¸
µ
¸
ÈÉ
¶tµ
°²±ÆÇÅ
񒏀
±ÂÁ
¸
µ
»Ê½
µ
5.2
¸
¶ÃµÄ
¸U¹¼»À¿
¶Ãµ
¸˜Ä
µ
Find the values for parameters
is controllable
,
ËÌ
, and
ËÍ
such that the given system
ËÎ
ÈÉ
µ
¸
ÈÉ
¶tµ
°²±ÆÅÇ
񒏀
¸
·
¸
¶J·
¸
Ï
Ë Ì
¸
¸
Ë7Í
Ë Î
¸
»;½
5.3
¸
»¾¿
Find the values for parameters
is observable
°¾±S³1¶tµ
µ
and
Ð*Ì
Ð7Í
µ
¶tµ
such that the following system
±ÒÁ
¹¼»Ñ¿
Ð Ì
Ð Í
Ä
254
CONTROLLABILITY AND OBSERVABILITY
If the output vector of the corresponding discrete system is given by
ÓYÔGÕ&Öv×ØÔ<ՐÙ8×ÚÛÜÓÝÙßÞ8Ú
, find the system’s initial condition.
5.4
Verify that all columns of the matrix
Ö
ÛÆãä
àâá
Ù
Þ
å
Ù
å
á
Ö
æ
Ù
Ö
Ù¼çè
can be expressed as a linear combination of the columns forming matrices
à
àêé
I, , and
(see 5.13).
5.5
Assuming that the desired final state of a discrete system represented by
Ö
Ûëãä
à
Ù
Þ
å
Ù
å
Õæv×{Û}ÓYÖ
is
ð
from
5.6
Ù
ÕÖD×
å
Õ&æv×
to
ð
ٝښò
Ö
Ö
æ
Ù
Ö
Ùìçèîí;ï
Ù
ÛÆãä Ù
Õ&Öv×ñÛÆãä Ù
ÞUçè”íØð
Ùìçè
find the control sequence that transfers the system
.
ð
Find a solution to Problem 5.5 in the case of a two-input system that has
the input matrix
Ö
ÛÆãä
ï
Ù
Ù
å
Ö
Ö
ÞUèç
The remaining elements are the same as in Problem 5.5.
5.7
Determine conditions on ó*ô ó é ô and
í
íaõ
í
is both controllable and observable
Ù
such that the following system
Ù
ÛSö
à
é
õ
Ö
ÛSö ó*ô
Ùì÷
íSï
Û}Ó
÷
ó é
íùø
õ
ô
Ú
é
õ
Assume that the input to this system is known. Find the initial conditions
of this system in terms of the given input in the case when the measured
Ô<Õú$×0ۖû8üìýþ7ÿ Õú$×
output is
.
5.8
Using the frequency domain criterion, check the joint controllability and
observability of the system
à
Ù
Û
ãä
Þ
å
Ö
Ö
Þ
֘çèJí"ï
Ù
Ö
Ù
Ù
ÛÆãä
å
Ö
ÛÂӜÙ
Ù4çè”í
ø
Ù
֘Ú
255
CONTROLLABILITY AND OBSERVABILITY
5.9
Find the initial conditions of all integrators in an analog computer simulation
of the following differential equation
5.10 The transfer function of a system given by
$
%#
##
!
&
'
%
%#
" #
&(
)
'
indicates that this system is either uncontrollable or unobservable. Check by
the rank test, after a zero-pole cancellation takes place, that the remaining
system is both controllable and observable.
5.11 A discrete model of a steam power system was considered in Mahmoud
(1982) and Gajić and Shen (1993).
(a) Using MATLAB, examine the controllability and observability of this
system, represented by
0/3 0/3
0/30
0/3
,- 0/210
=8>
'
54
'
64 >>
- 7 /8 ' 0/ 1
8
/
0
0
3
/
0
:
/
6
6
7
>
464
)9
0/
0/
0/30
0/3
*+ 7 /8 9
9)4
();
)
)4
. 7 / ; '
7 /8 (6(
7 /8 ( 0/ () 7 /8 () ?
0/361
0/3
7 /< )4
7 /8 7 /8 )
(69
0/3
0/
0/<
@ + BA2/80$ 0/:
((
9
'96(
'DCFE
G + IH
KJ
(b) Find the system transfer function and justify the answer obtained in (a).
5.12 Using MATLAB, examine the controllability of the magnetic tape control
system considered in Chow and Kokotović (1976)
,- ,- = >
0/ 0/36
0/36 = >
>
>
)
0/8
0/ 1
0/36
* @ )
/
.
. ?
7 ' / ()
7 ) / 96'
( 96( ?ML
0/8
0/36
N
7 $0/8
256
CONTROLLABILITY AND OBSERVABILITY
5.13 Find the controllability matrix of the system in the phase variable canonical
form and show that its rank is always equal to O .
5.14 Linearize the given system at the nominal point PQSRTVUQ0WXTYUZT0[]\^P`_6U%a0U%bS[
and examine system controllability and observability in terms of b
c
W
Q R \dQ R ZeU
Q R Pa[f\_
c
QSWg\hQSRQSWfiZeU
QSWjPa[k\la
m \dQ R
5.15 Find the state space form of a system given by
n6o m
n W m
n m
n
Z
m
np o i np W &
i b np irq \ np iZ
and examine system controllability and observability in terms of b . Do they
depend on the choice of the state space form?
5.16 Given a linear system described by
n W m
n m
n
Z
m
np W irq np i
\ np i&s
Z U
n m
P a[
np
\hq
m P#a[\ha0U
Transfer this differential equation into a state space form and determine the
initial conditions for the state space variables. Can you solve this problem
by using an unobservable state space form? Justify your answer.
5.17 Check that the matrix t given in (4.36) and the matrix uv\w x
in (4.39) form an observable pair.
defined
5.18 Derive formulas (5.55) and (5.56) for the reduced-order observer design.
5.19 Using MATLAB, examine the controllability of a fifth-order distillation
column considered in Petkov et al. (1986)
z{
8‡ ˆ
{ a:_N€N
ˆ
a8a6‚6qƒ
a
a
a
{
ˆ
{|~} _ „a6‚
ˆ
q0:_N„6q
a03€6ƒ6a6…
a
a
}
ty\
a
_ †€6†
„0:_€
_8†…
a
} q8‚6„q
a
a8a„6††
Y3q6†…
_63ƒ6†† ‰
} a0:_N‚6„‚
a
a03a6aq6q…
a
a<_N‚q6†
}
Š
\Œ‹
a
a
a03a6‚„6q
a
}
a8aƒ6„ƒ
a0:_N„€6‚
}
a0<_$a6a
a8q6a‚6a
}
a03a6a6‚„
a8a_Nq6ƒVŽ
257
CONTROLLABILITY AND OBSERVABILITY
5.20 Examine both the controllability and observability of the robotic manipulator acrobot whose state space matrices are given in Problem 3.2.
5.21 Repeat problem 5.20 for the industrial reactor defined in Problem 3.26.
5.22 Consider the state space model of the flexible beam given in Example 3.2.
Find the system transfer function and determine its poles and zeros. Use
Theorem 5.7 to check the controllability and observability of this linear
control system.
5.23 The system matrix for a linearized model of the inverted pendulum studied
in Section 1.6 is given in Section 4.2.3. Using the same data as in Section
4.2.3, the input matrix is obtained as
‘’8“
”
“
•M–˜—:™
Examine the controllability of this inverted pendulum.
5.24 A system matrix of a discrete-time model of an underwater vehicle is given
in Problem 4.22. Its input matrix is given by Longhi and Zulli (1995)
•g“08““6“–
“
š› “8“6–6ž6Ÿ
¡8¢
›
¢
› “0:”N“6–5 ¢
•g“08““0”$“
“
›
¢
› •M“8““6“0”
¢
“03“6–ž6Ÿ
“
œ
›
¢
‘
•M“8““6“6Ÿ
“0:”N“–6 “
“
“
“03“6“ž6ž £
“
“
“03“6––0”
Check the controllability of this system.
Download